Simple workflow for loop i2v animation.
Missing custom nodes can be installed by ComfyUI-Manager.
Based on parts from other workflows:
Wan 2.1 seamless loop workflow (i2v) - v1.0 | Wan Video Workflows | Civitai
Description
Use one of this models for Wan 2.2 workflow
FAQ
Comments (35)
wan 2.2 is 10/10 ♥️♥️
I wanted this so much... but i can't make it work, some nodes i can not get! any chance for a tutorial for dummies like me?
Error message occurred while importing the 'ComfyUI Upscaler TensorRT' module.
How to fix this please?
I just added a version with a default upscaller. You can try it. Same archive, but you will need to download it again.
@Discocat Thanks so much! It worked! Also I assume the Interpolation and Upscaler is OFF by default, right? Edit: I enabled them by using ctrl+b but after execution, the two boxes are highlighted in red. Any reason why?
2) What is the maximum limit I can increase the duration? My gpu is 4090.
@maaade Red borders in ComfyUI means error, you need to check it. Also use switches instead Ctrl + B. Max video duration depends on many things like your free RAM, GPU, amount of used lora's and model size. I am able to generate 3.5 sec video 1280p with 12 GB VRAM without lora's.
@Discocat Edit: Nevermind, thanks for the tip. I reloaded the workflow and used the switches to activate them and seems to work fine now :)
@Discocat Another thing, how does the speed scale matter? I see the default is at 5.
@maaade Its shift parameter for wan models. More value - more fast movements.
@Discocat I was trying some loras and they don't seem to work properly, especially this popular lora - https://civitai.com/models/1361346. It seems to work for everyone else but for me, the character doesn't walk despite using the correct prompts. Is it because of the looping nature of the workflow?
@maaade I dont know. This lora works fine for me. Here is expample with metadata included https://civitai.com/images/108833774
Try to add weight for lora's, or maybe another input image.
@Discocat Okay I was using smoothmix. Can you try it with that too?
@Discocat I noticed you changed the End Frame value to 2 while the one I originally downloaded in this page had "3" as its value.
2) Normalized Attention and CFGZeroStar also seems to have been disabled in your new example while the original WF had them active.
Any reason u changed these?
Edit: Used same settings except with Smoothmix and still same problem. So I am going to try the checkpoint you used. Maybe that might fix it.
@maaade You will need ajust settings for better results too. Like "Start frame" and "End frame" in "Frame overlap" group to get the smooth transition. This is the reason these settings were created.
@Discocat Ahh getting same result even with the other checkpoint. Now this is really mind boggling me because the lora is working fine in other wan workflows. Damn, this is the first time things aren't making sense.
Thank you for the workflow. I'll have a try, I'm just about to learn how this loop thing works. I wanted to interoperate this with my workflow as simple as possible. Maybe your wf is a good starting point. Your samples look very amazing :)
God damn what did you do to remove the color degradation over time? Been trying to figure this out for 2 weeks.
Actually nothing.This problem was solved by using checkpoints with merged lightx lora's (check suggested resources).
@Discocat Ohh that explains it! so weird how both seperated create this issue only for loops.
This guy created custom nodes to fix it with regular models, but only worked for 5b version
https://civitai.com/models/1931348
Only issue with these 2 models is that they have tons of additional loras, which seems to be hard to create an idle animation with little movements.
Can't select High noise and Low noise models, any solutions?
Download them in models/unet folder
@nothingherern450454 Can't select them, they are downloaded
@cockandballsyeh816 I have my models in "ComfyUI\models\diffusion_models" folder
@Discocat found out that the problem was that it doesn't detect gguf models
Best seamless loop workflow right now, imho.
I loved the 2.1 version of this, is there a way to use GGUFs instead ?
Yes. Replace model load nodes with "Unet Loader (GGUF)" and use this models (may not work with others).
If you encounter an error of "JWImageExtractFromBatch" node, the solution is here.stop working with update · Issue #20 · jamesWalker55/comfyui-various
Please tell me how to do it specifically. I don't have that Python directory. How should I proceed
Can someone tell me why the animation is always very very quick? i put the duration to 3s and the speed to 1 and 24 frames, but it's still giving me a very fast animation.. xd
try reduce the weight of the movement prompt by putting (prompt:0.4) ,and also lowering cfg scale will generally slow things down
its because of the 24 frames, every second is 24 frames so you maxed out the video in 1 second, do 8 fps and you'll hit your desired 3 seconds, or greatly increase the total amount of frames generated.
change NAG scales at this node. try speed = 10 (yes, its strange but higher amount = more smooth movements)
it keeps saying node 530 missing
Another WF that doesn't work even after installing the nodes manually. This ish is a joke.